Spatio-temporal Modeling Using a Cost-Effective System for Air Quality Monitoring

Abstract

Ambient air quality remains a major environmental and public-health concern. While traditional reference-grade monitoring stations provide reliable pollutant measurements, their high installation and maintenance costs limit spatial coverage—particularly in low-resource regions. Moreover, these stations generally do not provide future pollutant concentrations. To address these constraints, the South African Consortium for Air Quality Monitoring (SACAQM) has developed a cost-effective air quality monitoring system that integrates cost-effective sensors, advanced IoT technologies, and predictive modeling capabilities. AI r system provides real-time pollutant data across multiple locations in Gauteng province through a centralized dashboard. After calibration against reference-grade stations, the system demonstrated high reliability, highlighting its potential as a practical alternative to traditional monitoring infrastructure. This study presents the first use of spatio-temporal modeling for 24-hour forecasting using data from a cost-effective air-quality monitoring system deployed in Africa. We applied spatio-temporal deep learning models to pollutant data collected from several sites monitored by AI r system, enabling the extraction of both spatial and temporal patterns relevant for forecasting. The results illustrate the feasibility and promise of combining cost-effective sensing networks with spatio-temporal approaches for air-quality forecasting and early warning.